Font and Size Independent OCR for Printed Kannada Documents using SVM Classifiers

The following topics were dealt with: document analysis and recognition; multimedia document processing; character recognition; document image processing; cheque processing; form processing; music processing; document segmentation; electronic documents; character classification; handwritten character recognition; information retrieval; postal automation; font recognition; Indian language OCR; handwriting recognition; performance evaluation; graphics recognition; oriental character recognition; and word recognition

On improving the generalization of SVM classifier

The generalization performance of the SVM classifier depends mainly on the VC dimension and the dimensionality of the data. By reducing the VC dimension of the SVM classifier, its generalization performance is expected to increase. In the present paper, we argue that the VC dimension of SVM classifier can be reduced by applying bootstrapping and dimensionality reduction techniques. Experimental results showed that bootstrapping the original data and bootstrapping the projected (dimensionally reduced) data improved the performance of the SVM classifier.

Hybrid Working Set Algorithm for SVM Learning With a Kernel Coprocessor on FPGA

Support vector machines (SVM) are a popular class of supervised models in machine learning. The associated compute intensive learning algorithm limits their use in real-time applications. This paper presents a fully scalable architecture of a coprocessor, which can compute multiple rows of the kernel matrix in parallel. Further, we propose an extended variant of the popular decomposition technique, sequential minimal optimization, which we call hybrid working set (HWS) algorithm, to effectively utilize the benefits of cached kernel columns and the parallel computational power of the coprocessor. The coprocessor is implemented on Xilinx Virtex 7 field-programmable gate array-based VC707 board and achieves a speedup of upto 25x for kernel computation over single threaded computation on Intel Core i5. An application speedup of upto 15x over software implementation of LIBSVM and speedup of upto 23x over SVMLight is achieved using the HWS algorithm in unison with the coprocessor. The reduction in the number of iterations and sensitivity of the optimization time to variation in cache size using the HWS algorithm are also shown.

Detección de personas basada en HOG y SVM

El presente proyecto se centra en proteger a los peatones de las vías urbanas o de una fábrica donde conviven con robots móviles, pues son los mayores afectados en los acci- dentes producidos en estos entornos. El objetivo es diseñar un algoritmo basado en visión monocular capaz de detectar a los usuarios de forma rápida y precisa de tal forma que se tenga constancia en todo momento de los peatones que se encuentran delante del vehículo.

应用SVM预测杜鹃属（Rhododendron L.）在中国的潜在分布中心

杜鹃属(Rhododendron L.)是中国种子植物中最大的属，其现代分布和分化中心是我国西南部的横断山区和东喜马拉雅地区。我国西部、西南部的云南、四川、西藏等地共有杜鹃达450种，仅特有种就有约300种。对杜鹃属分布的深入研究是横断山区生物多样性保护不可缺少的重要部分。 由于物种分布与环境因子之间存在着紧密的联系，利用环境因子作为预测物种分布模型的变量是当前最普遍的建模思路。但是绝大多数物种分布预测模型都遇到了难以解决的“高维小样本”问题――模型在标本数据不足时无法给出合理的预测，或者模型无法处理大量的环境变量。机器学习领域的理论和实践已经证明，基于结构风险最小化原理的支持向量机(Support Vector Machine, SVM)算法非常适合“高维小样本”的分类问题。为了探索其应用在物种分布预测问题上的可能性，本文创新性的实现了基于SVM算法的物种分布预测系统。然后，本文以30个杜鹃属(Rhododendron L.)物种为检验对象，利用其标本数据和11个1km的栅格环境变量图层作为模型变量，预测其在中国的潜在分布区。本文通过全面的模型评估——专家评估，ROC (Receiver Operator Characteristic)曲线和曲线下方面积AUC (Area Under the Curve)——来比较模型的性能。试验结果表明，我们所实现的以SVM为核心的物种分布预测系统无论在计算速度还是预测效果上都远远优于当前广泛使用的GARP (Genetic Algorithm for Rule-Set Prediction)预测系统。 之后，本文进一步探讨了SVM预测系统预测效果与环境变量维数和标本点个数的关系。试验结果表明，对于只有少量标本点的物种SVM的预测结果仍然具有相当的合理性。由此可见， SVM预测系统很好的解决了以前众多模型无法克服的稀有种和标本点稀少的物种的潜在分布区模拟问题。同时本文发现大的环境维数（高维）对于物种潜在分布区的预测有着决定性的作用，因此模型处理高维问题的能力显得至关重要。 最后，我们使用中国所有可获取的杜鹃属标本数据，以及83个1km的栅格环境变量图层，对400种杜鹃属物种的潜在分布区进行预测。根据预测出来的物种潜在分布区，我们得到了中国杜鹃属物种潜在多样性分布格局，特有物种潜在多样性分布格局，濒危杜物种潜在的分布格局，各亚属物种潜在分布格局，以及不同生活型物种潜在多样性分布格局。这些分布区图不仅可以对杜鹃属起源研究提供分析验证的条件，还能为其引种、保护和新种的搜寻提供有利的空间依据。

QE and Suns-V-oc study on the epitaxial CSiTF solar cells

In order to clarify the major factors having confined the efficiencies of as-prepared crystalline silicon thin film (CSiTF) solar cells on the SSP (silicon sheets from powder) ribbons, QE (quantum efficiency) and Suns-V-oc study were performed on the epitaxial CSiTF solar cells fabricated on the SSP ribbons, the SSP ribbons after surface being zone melting recrystallized (ZMR) and single crystalline silicon (sc-Si) substrates. The results show that the epi-layers deposited on the SSP ribbons have rough surfaces, which not only increases the diffusion reflectance on the surfaces but also makes the anti-reflection coatings become structure-loosened, both of which would deteriorate the light trapping effect; in addition, the epi-layers deposited on the SSP ribbons possess poor crystallographic quality, so the heavy grain boundary (GB) recombination limits the diffusion length of the minority carriers in the epi-layers, which makes the as-prepared CSiTF solar cells suffer the worse spectra response at long-wavelength range. Nearly all the dark characteristic parameters of the CSiTF solar cells are far away from the ideal values. The performances of the CSiTF solar cells are especially affected by too high I-02 (the dark saturation current of space charge region) values and too low R-sh (parallel resistance) values. The higher 102 values are mainly caused by the heavy GB recombination resulting from the poor crystallographic qualities of the silicon active layers in the space charge regions, while the lower R-sh values are attributed to the electrical leakage at the un-passivated PN junction or solar cell edges after the solar cells are cut by the laser scriber.

The image coding scheme based on neural network and SVM

This paper applies data coding thought, which based on the virtual information source modeling put forward by the author, to propose the image coding (compression) scheme based on neural network and SVM. This scheme is composed by "the image coding (compression) scheme based oil SVM" embedded "the lossless data compression scheme based oil neural network". The experiments show that the scheme has high compression ratio under the slightly damages condition, partly solve the contradiction which 'high fidelity' and 'high compression ratio' cannot unify in image coding system.

Lossless data compression scheme based on neural network and SVM

First, the compression-awaited data are regarded Lis character strings which are produced by virtual information source mapping M. then the model of the virtual information source M is established by neural network and SVM. Last we construct a lossless data compression (coding) scheme based oil neural network and SVM with the model, an integer function and a SVM discriminant. The scheme differs from the old entropy coding (compressions) inwardly, and it can compress some data compressed by the old entropy coding.

A CMOS front-end circuit for SONET OC-96 receiver

An optical receiver front-end for SONET OC-96 receivers was analyzed and designed in 0.18 mu m CMOS process. It consists of a transimpedance amplifier (TIA) and a limiting amplifier (LA). The TIA takes a fully differential configuration, and regulated cascode (RGC) input stage is implemented. The LA was realized by five cascaded identical gain stages with active inductor load. The TIA achieved 4.2GHz bandwidth for 0.5pF photodiode (PD) capacitance and 1.2k 0 transimpedance gain. The LA achieved 5.4GHz bandwidth and 29dB voltage gain. The optical sensitivity is -19dBm at 5-Gb/s for a bit-error rate of 10(-12), and it dissipates 45.5mW for I.8V supply.

Comparing the Performance of Biomimetic Pattern Recognition with W8 and SVM on Prediction of Horizontal Gene Transfers in Bacteria Genomes

National Natural Science Foundation of China 60753001

基于模板匹配和SVM的草图符号自适应识别方法

草图符号的自适应学习中，不同用户的训练样本数量可能不同。保持在不同样本数量下良好的学习效果成为需要解决的一个重要问题．提出一种自适应的草图符号识别方法，该方法采用与训练样本个数相关的分类器组合策略将模板匹配方法和SVM统计分类方法进行了高效组合．它通过利用支持小样本学习的模板匹配方法和支持大量样本学习的SVM方法，并同时利用草图符号中的在线信息和离线信息，实现了不同样本个数下自适应的符号学习和识别．基于该方法，文中设计并实现了支持自适应识别的草图符号组件．最后，利用扩展的PIBGToolkit开发出原型系统IdeaNote．评估表明，该方法可以在24类草图符号分别使用1到20个训练样本时具有较高的识别正确率和较好的时间性能．